DynaRNA: accurate dynamic RNA conformation ensemble generation with diffusion model

RNA plays a wide variety of roles in biological processes. In addition to serving as the coding messenger RNA (mRNA), the vast majority of RNAs function as non-coding RNAs (ncRNAs), where their dynamic structural ensemble is critical for mediating diverse biological functions. However, traditional experimental techniques and molecular dynamics (MD) simulations face significant challenges in characterizing the conformational dynamics of RNA, due to inherent methodological limitations and high computational cost. We herein present DynaRNA, a diffusion-based generative model for RNA conformation ensemble. DynaRNA employs a denoising diffusion probabilistic model (DDPM) with an equivariant graph neural network (EGNN) to directly model RNA 3D coordinates, enabling rapid exploration of RNA conformational space. DynaRNA enables end-to-end generation of RNA conformation ensemble reproducing experimental geometries without the need for Multiple Sequence Alignments (MSA) information. Our results demonstrate that DynaRNA effectively and accurately generates a tetranucleotide ensemble with a lower intercalation rate than molecular dynamics simulations. Besides, DynaRNA has the ability to capture the rare excited state of HIV-1 Trans-Activation Response (TAR), and recapitulate de novo folding of tetraloops. DynaRNA can serve as a complementary tool with current computational tools, such as molecular dynamics simulations, and a versatile and efficient platform for modeling RNA structural dynamics with broad implications and potential in RNA structural biology, synthetic biology, and therapeutic development.